← Back
OpenAI
OpenAI releases GPT-5.4 mini in ChatGPT; adds mid-response course correction to reasoning model
OpenAI APIChatGPTOpenAI · releasefeaturemodelapideprecation · help.openai.com ↗

GPT-5.4 Mini Rollout

OpenAI is rolling out GPT-5.4 mini across ChatGPT tiers with different availability levels. Free and Go users can access it directly via the "Thinking" feature in the + menu, while Plus, Pro, and other paid users will see it deployed as a rate-limit fallback for GPT-5.4 Thinking when usage caps are reached. Enterprise customers retain the option to set Auto routing to default to GPT-5.4 mini. The model will not appear as a selectable option in the model picker.

GPT-5.4 Thinking Improvements

GPT-5.4 Thinking brings major advances in reasoning and coding capabilities, incorporating technology from GPT-5.3-Codex while improving multi-tool workflows. The key innovation is the ability to adjust course mid-response—users can now see the model's upfront thinking plan and steer it before it reaches a final output, reducing the back-and-forth needed to get desired results.

Additional improvements include:

  • Enhanced web research for highly specific queries
  • Better context management for longer thinking sessions
  • Faster reasoning without sacrificing depth
  • Improved performance on complex professional tasks involving spreadsheets, presentations, and documents

GPT-5.3 Instant & Legacy Model Changes

GPT-5.3 Instant received an update reducing teaser-style phrasing ("You'll never believe…") and improving follow-up tone. The model now delivers more measured, grounded responses with better-contextualized web search results.

Model Deprecations:

  • GPT-5.1 models (Instant, Thinking, Pro) are retired as of March 11, 2026
  • GPT-5.3-Codex launched as OpenAI's most capable agentic coding model, running ~25% faster with improved benchmark performance
  • Previously announced: GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini retired from ChatGPT

Developer Implications

For ChatGPT users: Model availability has shifted; ensure custom workflows account for the fallback behavior of mini models during high usage periods. The mid-response steering feature in GPT-5.4 Thinking enables more iterative problem-solving workflows.

For API users: No API changes announced at this time, though GPT-5.3-Codex represents a new option for advanced coding tasks requiring agentic behavior.